Supporting data for “Reconstructing 3D Indoor Scene From RGB Equirectangular Panorama Image with CNN System”

This dataset includes 3DOF pose (Azimuth, Elevation, In-plane rotation) for objects in indoor environments. We have 10 categories, which are mostly common household furniture. We collected images from ImageNet, Google, Baidu and Bing, and 3D CAD models from Trimble 3D Warehouse. We created an annotation tool for annotating 3D Poses. Specially, annotators can identify objects inside images, specify them with bounding box, and select a corresponding CAD model for the object. Then annotators can align 3D pose of CAD model against the object in the image, and save the 3D pose as annotation files. There are a total of 3 folders in the dataset. Annotations, which is all the annotated data in .json; Build24_5, which is the annotation tool; and Images, which contains all images used in our dataset.